Study On High Resolution Target Detection And Information Transmission Based On Compressed Sensing

With rapid growth in demand for objective information in military and civilian areas, how to effectively detect target information have been studied extensively. However, with the development of high resolution techniques, high sampling rate determined by the Nyquist theory brings a great challenge of acquiring. Furthermore, the resultant huge amounts of data also give enormous pressure to data storage and transmission within limited bandwidth channel.The recently introduced compressed sensing (CS) theory indicated that, by solving an nonlinear optimization problem, a sparse or compressible signal can be reconstructed from incomplete measurements with overwhelm probability. In general, the signal is projected from a high-dimensional space into a low-dimensional space and it greatly reduces the sample rate. Inspired by CS theory, by exploring the sparsity of the signal, a new information transmission scheme and robust target detection methods are investigated in this dissertation. The content includes three aspects, e.g., one-dimensional parameter estimation (angle, range), two-dimensional radar imaging and remote sensing image transmission, and the contributions are taken as follows.(1) By employing the generalized Cauchy distributions as the prior of the signal, we propose a compressed sensing reconstruction algorithmand apply it into the radar angle and range estimation. Specifically, with the sparsity enhancement property of generalized Cauchy distributions, a associated reconstruction model by estimating the maximum a posteriori of the signalin a Bayesian framework is developed. Due to its non-convex nature, the model is a non-convex minimization problem. In this dissertation, by introducing a weighting mechanism, the non-convex optimization problem is converted into convex sub-optimization problems which greatly simplifies the computational complexity. Finally, the proposed method is applied into ultra-wideband radar range and direction-of-arrival (DOA) estimation, and the experimental results show that, high resolution performance in range and DOA estimation can be achieved when in the low sampling measurements case.(2) For a SAR scene which is rich in spatial texture, the compressive SAR imaging with fixed sparse representation cannot achieve the superior reconstruction. To solve this problem, we propose a novel structural sparse representation based SAR imaging approach, in which a structural space derived from piecewise autoregressive model is constructed for adaptive image sparse representation. Also, similarity among pixels is incorporated into the piecewise autoregressive model to further illustrate the local image non-stationary property. Since the piecewise autoregressive model is inherently determined by the unknown image, a joint optimization scheme is used in the algorithm by iterative SAR imaging and updating of the structural space. Eventually, experimental results demonstrated that the proposed algorithm has better SAR reconstructed images then conventional ones.(3) Near Space communication between hypersonic aircraft and ground stations suffers from unstable communication within limited bandwidth. To solve this problem, a robust Multiple-Stations Scalable Transmission scheme based on CS is proposed. The proposed scheme exhibits two advantages. Firstly, multiple-station receivers are disposed at different locations to accomplish continuous and real-time communication with hypersonic vehicle. Secondly, by exploring the correlation among measurements, a scalable CS-based coding method is adopted in which a two-layer structure is used to improve coding efficiency. Due to the equal importance of CS measurements, the proposed scheme is capable of eliminating the influence of data loss caused by communication interrupt. Experimental results show that the proposed scheme can achieve a reliable communication when the data loss is up to 30%. Furthermore, we propose a fast reconstruction algorithm for lp norm non-convex model and apply it into multiple descriptions coding. To reduce storage pressure, a fast gradient descent method, whose matrix inverse operations can be implemented implicitly, is applied and a novel weighted norm method is used to optimize the descent steps, which greatly improves the algorithm speed. Experimental results show that the proposed algorithm provides a high computational efficiency.